[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <50EB0278.4010806@free-electrons.com>
Date: Mon, 07 Jan 2013 18:14:32 +0100
From: Gregory CLEMENT <gregory.clement@...e-electrons.com>
To: Thomas Petazzoni <thomas.petazzoni@...e-electrons.com>
CC: "David S. Miller" <davem@...emloft.net>,
Lior Amsalem <alior@...vell.com>, Andrew Lunn <andrew@...n.ch>,
Jason Cooper <jason@...edaemon.net>, netdev@...r.kernel.org,
Maen Suleiman <maen@...vell.com>,
Dmitri Epshtein <dima@...vell.com>,
linux-arm-kernel@...ts.infradead.org
Subject: Re: [PATCH 3.8-rc] net: mvneta: fix driver operation in SMP context
On 01/07/2013 05:27 PM, Thomas Petazzoni wrote:
> From: Dmitri Epshtein <dima@...vell.com>
>
> In order for the driver to behave properly in a SMP context, the same
> transmit queue should be used by the kernel in dev_queue_xmit() and in
> the driver's mvneta_tx() function. To achieve that, the driver now
> implements the ->ndo_select_txq() operation.
>
> For now, it always returns the same transmit queue, txq_def, until the
> driver is expanded to properly take advantage of the multiqueue
> capabilities of the hardware.
>
> Without this patch, the network driver crashes the kernel almost
> immediately on Armada XP platforms, if the network load is at least a
> little bit parallel (i.e several threads).
>
> [Thomas Petazzoni: reword commit message]
> Signed-off-by: Dmitri Epshtein <dima@...vell.com>
> Signed-off-by: Thomas Petazzoni <thomas.petazzoni@...e-electrons.com>
I have tested this patch on my Armada XP DB board and on the
OpenBlocks AX3 board. I confirmed that it fixes the problem when
multiple thread are used.
You can add my
Tested-by: Gregory CLEMENT <gregory.clement@...e-electrons.com>
and also my
Acked-by: Gregory CLEMENT <gregory.clement@...e-electrons.com>
Or a combination of the two if you prefer!
> ---
> This is 3.8-rc material.
> ---
> drivers/net/ethernet/marvell/mvneta.c | 17 +++++++++++++++--
> 1 file changed, 15 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/net/ethernet/marvell/mvneta.c b/drivers/net/ethernet/marvell/mvneta.c
> index b6025c3..af2c421 100644
> --- a/drivers/net/ethernet/marvell/mvneta.c
> +++ b/drivers/net/ethernet/marvell/mvneta.c
> @@ -1310,6 +1310,17 @@ static u32 mvneta_skb_tx_csum(struct mvneta_port *pp, struct sk_buff *skb)
> return MVNETA_TX_L4_CSUM_NOT;
> }
>
> +static u16 mvneta_tx_policy(struct mvneta_port *pp, struct sk_buff *skb)
> +{
> + return (u16)txq_def;
> +}
> +
> +static u16 mvneta_select_txq(struct net_device *dev, struct sk_buff *skb)
> +{
> + struct mvneta_port *pp = netdev_priv(dev);
> + return mvneta_tx_policy(pp, skb);
> +}
> +
> /* Returns rx queue pointer (find last set bit) according to causeRxTx
> * value
> */
> @@ -1476,7 +1487,8 @@ error:
> static int mvneta_tx(struct sk_buff *skb, struct net_device *dev)
> {
> struct mvneta_port *pp = netdev_priv(dev);
> - struct mvneta_tx_queue *txq = &pp->txqs[txq_def];
> + u16 txq_id = mvneta_tx_policy(pp, skb);
> + struct mvneta_tx_queue *txq = &pp->txqs[txq_id];
> struct mvneta_tx_desc *tx_desc;
> struct netdev_queue *nq;
> int frags = 0;
> @@ -1486,7 +1498,7 @@ static int mvneta_tx(struct sk_buff *skb, struct net_device *dev)
> goto out;
>
> frags = skb_shinfo(skb)->nr_frags + 1;
> - nq = netdev_get_tx_queue(dev, txq_def);
> + nq = netdev_get_tx_queue(dev, txq_id);
>
> /* Get a descriptor for the first part of the packet */
> tx_desc = mvneta_txq_next_desc_get(txq);
> @@ -2550,6 +2562,7 @@ static const struct net_device_ops mvneta_netdev_ops = {
> .ndo_change_mtu = mvneta_change_mtu,
> .ndo_tx_timeout = mvneta_tx_timeout,
> .ndo_get_stats64 = mvneta_get_stats64,
> + .ndo_select_queue = mvneta_select_txq,
> };
>
> const struct ethtool_ops mvneta_eth_tool_ops = {
>
--
Gregory Clement, Free Electrons
Kernel, drivers, real-time and embedded Linux
development, consulting, training and support.
http://free-electrons.com
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Powered by blists - more mailing lists